Floating-Point Error Handling in C++: What Actually Works